Abstract: A new method was proposed to predict earthquake through using generalized regression neural network(GRNN) based on genetic algorithm(GA).Global optimum can be found through using GA's ability of global searching and being not liable to get the minimum to optimize the the center and width of radial basis function and weights of output layer.At last, using 100 sample data obtained from actual earthquake, a simulation experiment is done based on MATLAB.The simulation shows that the method has a good result.